The history is confusing. bpo user Louie Lu (louielu), as github user 'noname louisom', open PR 1511 for this existing issue on 2017 May 9.  On May 12, he opened #30348 and PR 1512 with the tests for fetch_completions and get_entity that were part of PR 1511.  (This was a needed separation.)

By June, he had switched to a new github name 'Louie Lu mlouielu'.  On June 12, he opened #30632 and PR 2124, which duplicated #30348 and PR 1512 (which partly duplicated PR 1511).  On June 15, he closed PR 1511 to 'migrate' it to pr 2209.  But the latter only included the tests also in PR 1512, which it replaced on #30348, and PR 2124.  He also closed #1512.  Loiue moved on to other projects in Fall, 2017.

After revisions, I merged PR 2209 for #30348 last March.  I followed up with #36419, PR 15121, now merged. I just opened #37766 to finish preparing autocomplete for new additions such as this.  I was thinking of this issue when I included adding an htest.  (Note: #27609 is the master issue for completions.)

Notes for migrating the dict keys code:
1. In PR 15121, I shrank mode names to ATTRS and FILES.  The new mode name should be KEYS, or maybe SKEYS (for string keys).  Other refactors should not affect KEYS too much.
2. I intend to change 'smalll' and 'bigl' to 'small' and 'big' and might make other changes to fetch_completions.
3. I intend to split test_fetch_completions into separate methods for each mode. The new KEYS tests should be a separate method 'test_fetch_keys'.

The questions of function calls in the entity expression is more nuanced than I know when I wrote msg293520.  For force-open-completions, control-space, function calls are allowed.  But I do not think that this affects the new mode.